How Spatial Is Hyperspace? Interacting with Hypertext Documents: Cognitive Processes and Concepts

CYBERPSYCHOLOGY & BEHAVIOR Volume 4, Number 1, 2001 Mary Ann Liebert, Inc.

PATRICIA M. BOECHLER

ABSTRACT

The World Wide Web provides us with a widely accessible technology, fast access to massive amounts of information and services, and the opportunity for personal interaction with numerous individuals simultaneously. Underlying and influencing all of these activities is our basic conceptualization of this new environment; an environment we can view as having a cognitive component (hyperspace) and a social component (cyberspace). This review argues that cognitive psychologists have a key role to play in the identification and analysis of how the processes of the mind interact with the Web. The body of literature on cognitive processes provides us with knowledge about spatial perceptions, strategies for navigation in space, memory functions and limitations, and the formation of mental representations of environments. Researchers of human cognition can offer established methodologies and conceptual frameworks toward investigation of the cognitions involved in the use of electronic environments like the Web.

INTRODUCTION only a new cognitive environment (“hyperspace”) where information is perceived, stored, OR CENTURIES, TEXT-BASED LITERATURE has manipulated, and retrieved in new ways, but Fbeen the primary mode for disseminating also a new social environment (“cyberspace”), and receiving information. The cognitive where one individual’s cognitions about this processes involved in reading and writing new environment interact with those of other books have been studied extensively for people who share it. Unlike physical space, the decades as they have played a crucial role in World Wide Web is not constrained by dis- the development of humankind. In the past two tance. Interactions between individuals can be decades, hypertext systems (computer-based numerous and instantaneous. The possibilities information systems of interconnected docu- for enhancing individuals’ lives through easily ments) have become increasingly pervasive in accessed personal interactions and the avail- work, home, and educational environments. ability of online services are abundant. With the advent of the World Wide Web, these The study of hypertext systems is a multi- information systems are heavily influencing disciplinary venture. Psychologists, computer how we develop as a society. However, unlike scientists, and educators all bring different ap- traditional text information sources, the World proaches and goals to the effort. Cognitive psy- Wide Web is interactive. This has created not chologists wish to know how humans concep-

Psychology Department, University of Alberta, Edmonton, Alberta, Canada.

23 24 BOECHLER tualize hyperspace and the nature of the men- performance, and spatial processing. Under- tal representations they form of hyperspace. standing the cognitive aspects of hyperspace Social psychologists want to understand how will help us to develop optimal technologies this new way of communicating impacts an in- that will guide the way we use the social envi- dividual’s sense of self, their social sphere, the ronment of cyberspace. Understanding the cog- group dynamics of this new environment, and nitive processes that shape many different in- the effect on society at large in both a regional dividuals’ perceptions of hyperspace will and global sense. Clinical psychologists are in- facilitate the creation of technologies that are vestigating the ways that these systems can be accessible to a wide range of people with a used to provide psychological services. Com- wide range of characteristics (e.g., age differ- puter scientists and software developers want ences) and skills (e.g., level of literacy). For the to know how to design an optimal interface for uninitiated, traversing the World Wide Web in users. Educators are interested in how this search of solutions to everyday problems can technology can be applied to enhance learning. be extremely daunting and can often end in This can be advantageous as a broad knowl- failure to locate useful information or access edge base and diverse perspectives are brought needed services. Providing navigation tools to bear on the topic. However, it also results in that are based on the cognitive processes of the lack of a unified theoretical foundation for users will do much toward making the World hypertext. Relative to many other areas of psy- Wide Web an accessible resource for everyone. chological study, research on the cognitive as- Psychological processes are often described pects of using such systems is in its infancy; via metaphors.3,4 Hyperspace is a dynamic, in- models and paradigms are still being devel- teractive space requiring new psychological oped. metaphors, new ways of thinking and inter- One point to note is that the above-mentioned preting information. However, before we can endeavors are not as distinct as one might ini- devise new metaphors that portray this envi- tially suppose. Basic cognitive processes (e.g., ronment accurately, it is prudent to closely ex- memory, perception, etc.) are the underpin- amine previous metaphors to identify benefi- nings for more complex processes (e.g., social aspects that can act as a foundation for new cial–emotional responses to our environment). metaphors. In the real world, we’ve identified many rela- One overarching metaphor seems to domi- tionships between basic cognitions and social nate the hypertext literature: the spatial responses.1,2 There are numerous instances of metaphor. The spatial metaphor arose out of a our perceptions of physical space producing need for a common language to discuss hy- psychosocial consequences (e.g., personal space pertext issues and a framework within which being violated, overcrowding, feeling lost). It is to develop usable interfaces. Its basic premise reasonable to suppose that how we perceive is that locating information in hyperspace has hyperspace may affect our social interactions similar psychological features to navigating in within it. Understanding our cognitive reac- physical space. Therefore, promoting mental tions to hyperspace may well aid us in pre- representations of spatial layouts of informa- dicting some of our social and emotional re- tion is a good approach to improving the hy- sponses to this new environment. To fully take pertext user’s ability to access information.5,6 advantage of the opportunities offered by cy- Navigational tools modeled after real-world berspace, we need to comprehend clearly how navigational aids such as maps appear to en- individuals take in information from hyper- hance performance on some measures7–11 but space. not on others.12–16 Correlational data on the re- To this end, the traditional paradigms and lationship between spatial ability and perfor- theories of cognitive psychology have much to mance has shown increases in speed of perfor- offer, providing us with established method- mance with higher spatial ability,17,18 but ologies and conceptual frameworks for explor- marginal effects on the accuracy of perfor- ing different facets of cognition relevant to hy- mance.19 Therefore, the empirical evidence to pertext navigation, such as memory, task date has not resolved the issue of how or to HOW SPATIAL IS HYPERSPACE? 25 what degree spatial processing is involved in uments that contain images, audio, and video the efficiency and effectiveness of hypertext clips are referred to as “hypermedia” docu- use. ments. In the hypermedia literature the terms The spatial metaphor seems easily under- “hypertext” and “hypermedia” are often used stood by novice users. However, if it is not an interchangeably. 20 accurate metaphor for the mental processes Each hypertext window contains icons that that humans engage in as they interact with hy- can be pictorial symbols or highlighted words pertext, or if we do not understand the aspects (hotwords). A window can contain any num- of the comparison that make the metaphor ben- ber of these icons. The icon represents the link eficial, then as more and more complex, pow- to the database for that particular piece of in- erful systems are developed this metaphor will formation. When clicked on with a pointing de- eventually be less useful. This would be so for vice (usually a mouse), the icon activates the both the goal of achieving optimal access to in- link to the referenced node in the database and formation and the goal of presenting informa- a new window is instantaneously opened con- tion in a form that can enhance social interac- taining the information in that node. Nodes tion or learning. that hold related information are linked. Some The following review draws from many di- nodes will have a multitude of links, both to verse areas of cognitive psychological study to and from other nodes based on their related- bring forth questions whose answers could ness, whereas other nodes may only have in- greatly impact the benefits we are able to gar- coming links. It is important to stress that links ner from cyberspace. It is an in-depth look at are not merely connections between nodes but the possible ways we can view the new envi- that they should specify the semantic relation- ronment of the World Wide Web. This may or ships between nodes. The author of the hyper- may not be as a space, therefore, this review text document is the engineer of the links based also critically examines the spatial metaphor on his or her semantic network; that is, his or for hypertext interaction. her conceptions of the interconnectedness of ideas within the chosen material.21,22 Links can be organized in two ways: in a hierarchical structure or in a network structure. AN INTRODUCTION TO HYPERTEXT In a hierarchical structure, nodes are only linked to the superordinate information above What is hypertext? them or the subordinate information below Generally, a hypertext system is an electronic them.21,23 In a network structure, any node can system of interconnected units of information be linked to any other. Multiple links can exist or text nodes. Many different types of hyper- between superordinate and subordinate infor- text systems have been designed, each consist- mation.23 Regardless of the specific structure, ing of its own specific structures that support because of its general node/link relational the connections (links) between the user inter- structure, a hypertext system is nonlinear24; face (the windows or pages a user views on the that is, unlike the pages in a book, users can ac- display terminal) and the underlying informa- cess different windows in different orders. tion system (database of text nodes). The user There is no predetermined sequential organi- interface is sometimes referred to as the “front zation of the underlying nodes. Typically, a hy- end” of a hypertext system and the underlying pertext system will also have a backtrack func- database as the “back end” of the system. In tion to allow the user to access previously this paper, the term “window” will be used in viewed windows.25 reference to the screen image a user may be viewing and “node” in reference to the infor- A brief history of hypertext systems mational node that the window is connected to in the underlying database. The computer en- The concept of hypertext was first outlined vironment of hypertext documents has come to by Vannevar Bush in 1945 with his Memex in- be referred to as “hyperspace.” Hypertext doc- formational system.24 This was a mechanical 26 BOECHLER system rather than an electronic system. It con- links and windows rather than through some sisted of several microfilm projection units on central, and cumbersome, change in the system which the viewer could display different structure. The term “hypertext” was imple- frames of microfilm. The viewer could con- mented by Nelson in 1965.24 struct links, which Bush called trails, between Since the initial implementation of these sys- two items by viewing the two frames simulta- tems, a multitude of hypertext systems have neously, then inserting code that denoted the been developed, each with its own purpose and names of the frames and their current positions specialized features. Systems have been devel- within each document. The important feature oped to address very definite informational of the Memex was that any item within the sys- purposes, and general hypertext systems have tem could be made to immediately and auto- been devised for the purpose of studying hy- matically access any other item. The key idea pertext itself as a technology. was that two items could be linked together for The variations of function and features fast and flexible access. The system itself was within different hypertext systems are so nu- never implemented, but Bush had the foresight merous and complex as to be well beyond the to envision a time when his idea could be ap- scope of this paper. What is important to note plied to computer systems, which at the time is that as these systems evolved, grew in size were in the early stages of development, inac- and intricacy, and became accessible to the gen- cessible and costly. eral public (culminating in the early 1990s with Since that time, many versions of hypertext the advent of the World Wide Web), the ne- systems have been developed and continue to cessity for understanding how humans interact be developed. Each system has its own specific with hypertext, and perceive and manipulate features, but Bush’s initial idea is still the most information from the system, became increas- essential feature; that the system consists of a ingly important. viewer connected to underlying informational To summarize, the defining features of hy- nodes that can be accessed with speed and flex- pertext are: a large collection of informational ibility by the user. nodes connected via machine-supported links, In 1963, Engelbart devised the oN Line Sys- which support massive cross-referencing and tem (NLS), with features we would easily rec- easy access. The term “hypertext” will be used ognize as attributes of current hypertext sys- as a general term for systems that entail these tems.24 The NLS highlighted three particular features regardless of their individual differ- aspects of hypertext: “a database of non-linear ences. text, view filters which select information from this database, and an interface which struc- Goals in hypertext use tured the display of this information for the terminal.” (p. 22).24 The accessibility of high-res- Different hypertext users have different olution displays prompted the focus on goals, which produce different navigational graphical displays of nodes (the single page strategies. Users may be searching out a very display we are so familiar with now) and the specific piece of information that they have alone-to-one correspondence of node to window. ready accessed, looking for a specific piece of Engelbart also devised the “mouse.” Engel- information that they do not know how to find, bart’s system focused on the front end of a hy- or they may be exploring the system with no pertext system, the user interface. specific informational goal. Conklin24 outlines Concurrently, Ted Nelson was developing a the three methods of conducting an informa- UNIX-implemented system that focused on the tional search in a hypertext system: (1) the user back end of a hypertext system, the database can visit a set of related windows or pages server. He called this system “Xanadu.” The through the traversal of links; (2) the user can goal of the Xanadu project was to develop a navigate through the document using a graph- system capable of storing and retrieving a mas- ical browser, which provides a graphical rep- sive corpus of literary works, a system that was resentation of the network structure; or (3) the easily expandable through the inclusion of new user can activate a queried search where the HOW SPATIAL IS HYPERSPACE? 27 system searches all documents to find any in- being responsible for decisions about which in- stance of a keyword or key phrase that is en- formation to access (which links to select) and tered. The query method is necessary for doc- in what order. To this end, the cognitive re- uments or networks of documents, such as the sources of the user must be allocated toward World Wide Web, that cannot be exhaustively several goals concurrently. The term “cognitive examined because they contain too many overhead” has been used to refer to the amount nodes. of cognitive resources necessary to successfully complete an informational task in hypertext.24 Advantages and disadvantages of hypertext With hypertext, users must evaluate information currently viewed, must recall information The main advantage of hypertext over tradi- previously viewed as well as its location relational text is that it is a high-speed method of tive to other viewings, and must create a plan accessing vast quantities of information. As to navigate toward desirable information (not mentioned earlier, there is no predetermined only deciding which windows to access but sequential organization of the underlying also which tools to use to execute the plan). nodes. This provides an increased flexibility in Kim and Hirtle6 label these cognitive tasks as the ways that users can experience the infor- (1) navigational tasks: planning and executing mation in the document compared to tradi- routes through the network; (2) informational tional text-based presentations. The user is free tasks: reading and understanding the contents to make link connections in any order based on presented in the nodes and their relationships, their own focus and interpretations of the ma- for summary and analysis; and (3) task man- terial, toward their own end goal. However, agement: coordinating informational and nav- this same feature that provides such individual igational tasks. The cognitive demands that freedom for the user also comes with costs. For these tasks require are compounded by in- example, at any point in time, the user only sees creasingly complex and voluminous hypertext the current window and the links represented structures.22–24,27,28 Degraded performance oc- by icons or highlighted text, leading from that curs when a hypertext user’s cognitive re- window to other related windows. This means sources are overwhelmed by these demands. the user must remember the windows already viewed and hypothesize about the overall size, Disorientation. One of the symptoms of cog- structure, and complexity of the entire hyper- nitive overhead is disorientation.24,28,29 It is text network, as it can never be viewed in its well documented that hypertext users will entirety. often report “feeling lost” when perform- Because of this, there are two related cate- ing poorly on an informational search.22,26,29,30 gories of problems associated with using hy- Based on subjective reports, Edwards and pertext systems: cognitive overhead and dis- Hardman30 characterized this phenomenon via orientation. three categories of the users’ experience: (1) the user does not know where to go next; (2) the Cognitive overhead. Unlike traditional text, user knows where to go but not how to get hypertext does not have conventional cues that there; (3) the user does not know where they lead the reader through a document (e.g., chap- are in relation to the overall structure of the ters, sections).26 With traditional text docu- document. Based on empirical measures, Foss31 ments, format often indicates the type and pur- proposed three types of disorientation prob- pose of a document and gives the reader hints lems and their symptoms: (1) Navigational dis- about where important information will be orientation problems, caused by a lack of found (e.g., a legal contract is easily distin- knowledge of organization and extent of the guished from a grocery list). In hypertext, no document as well as unfamiliarity with navi- such conventions have been established. In ad- gational tools. The symptoms of navigational dition, there are multiple routes or paths to ac- disorientation are looping, inefficient pathways cess a specific piece of information in the net- and query failures. (2) The embedded digres- work. This increased flexibility results in users sion problem caused by high cognitive de- 28 BOECHLER mands that lead to difficulties in planning, firmly entrenched the spatial metaphor is in the managing and executing digressions. The hypertext literature. The use of terms related to symptoms for embedded digression are a dis- space is solidly embedded in the topic of hy- organized screen layout, having many win- pertext. Terms such as “hyperspace,” “envi- dows open concurrently and engaging in ex- ronment,” and “navigation” are difficult to cessive backtracking. (3) The art museum avoid as, at this point in time, they provide for problem caused by the high cognitive demand the user the most direct access to meaning at of viewing many windows superficially with- an introductory level. To this end, the spatial out attention to detail or relatedness. The metaphor has been useful and may continue to symptoms of the art museum problem are short be useful. However, beyond the introductory reading times and restricted search paths. level, is the spatial metaphor impeding the de- The problems of disorientation and cognitive velopment of a more cognitively accurate overhead were identified by early hypertext re- model, one that might ultimately provide more searchers,22,23,29,30 but continue to be unre- benefit to the user? solved.13,28 It has been suggested that unless Before addressing this larger issue, it is pru- disorientation and cognitive overhead can be dent to attempt to make explicit the ways in effectively reduced the ultimate usefulness of which the terms “spatial” and “metaphor” are hypertext will be severely limited.24,28,29 used (or not used) in relation to hypertext.

The spatial metaphor WHAT IS THE ROLE OF A METAPHOR? The identification and labeling of the two problems of disorientation and cognitive over- The research on metaphor and its related head are a direct result of how we have come term, analogy, is vast and complicated. The fol- to conceive of the environment that hypertext lowing section is a brief review of perceptions creates. Disorientation is a cognitive phenom- of metaphor in both the psychological and hu- enon that occurs in an environment where it is man–computer interaction literature. necessary to maintain a sense of one’s own lo- In the psychological literature, Lakoff37 de- cation relative to the location of other objects scribes metaphor as a mode of thought that can and to maintain the relative location of these be understood as a tightly structured mapping objects to each other. Cognitive overhead in hy- from the source domain to the target domain. pertext is the amount of cognitive resources The source domain is a conceptual schema that that are required to successfully interact with we have constructed based on our experiences the type and number of structures within that in a particular environment. In the case of the environment (the type of structures dictated by spatial metaphor, the source domain is physi- our conception of the environment). cal space and the target domain is hyperspace. The spatial metaphor arose out of a need for Holyoak and Thagard3 consider metaphor, a common language to converse about the hy- like analogy, a method toward understanding pertext environment and, subsequent to that, out a new domain in terms of a source domain, but of the need for a framework for designing tools metaphor involves connections that move and hypertext features that create optimal navi- beyond surface category structure. “Many gating environments. Canter et al.5 have been at- metaphors are based on deeper relational and tributed with the emergence of the metaphor13 system mappings” (p. 217).3 which has been advocated by a number of Holyoak stresses the importance of the eval- researchers.6,8,9,15,32 However, other authors uation of these mappings based on similarity, have expressed concern over the rapid and seem- structure, and purpose between the source and ingly overwhelming acceptance of the meta- target domains. phor.13,33–36 According to Pylyshyn4 the role of a meta- Having presented an overview of hypertext phor for scientific explanation is not to fully ex- systems, and a brief description of how users plain all aspects of a complex environment but interact with them, it should be clear how to provide a framework on which to reference HOW SPATIAL IS HYPERSPACE? 29 new, vague, and disconnected ideas about that has been prematurely accepted as a theoreti- environment, phenomenon, or concept. This cally plausible metaphor for the cognitive framework is a means toward the development processes involved in hypertext interaction. of a more thorough understanding of something Unlike psychological theory, in the hu- new. Pylyshyn describes the acquisition of new man–computer interaction (HCI) literature the knowledge using Piaget’s terms of accommoda- end goal is to provide a metaphor for the user tion and assimilation. When we first encounter that helps them understand the purpose and new information, we attend to aspects that are function of the facilities of a specific applica- most similar to the schemas we’ve already con- tion, rather than to provide the user with an ex- structed and categorize the new information as planation of the global environment of hyper- belonging to the most similar schema (assimila- text. Metaphors can be applied to an tion). If the new information is so dissimilar that application with a narrow purpose (e.g., the it cannot be mapped onto an existing schema, we typewriter metaphor for word-processing) or a then adjust and expand the structure of the broader purpose (e.g., the general desktop schema to include this new knowledge (accom- metaphor used in many types of applications). modation). However, Pylyshyn maintains that it In the HCI literature, some authors are careful is important to provide some degree of explana- to explicitly present the spatial metaphor as a tion via the metaphor (a powerful metaphor) design tool for a specific application. An ex- rather than merely description, which may leave ample of this is Hammond and Allinson’s32 the user with a groundless impression of having travel holiday metaphor that includes choices been provided with an explanation (an impotent for go-it-alone travel or guided tour travel for metaphor). The value of a metaphor decreases if a specific hypertext system. Other authors pre- the user, upon closer inspection, realizes that the sent the spatial metaphor as if they accept the metaphor has few shared properties with the assumption that it’s a psychologically plausi- new situation the person is attempting to un- ble account of how people think about hyper- derstand. Surface similarities may prove of lim- space in general.6,8 ited value as functional and conceptual dissimi- Interface designers expect and accept the pres- larities are uncovered. ence of metaphorical mismatches. Carroll et al.38 In hypertext, as the user becomes more profi- emphasize the usefulness of mismatches as do cient the comparison between hyperspace and Hammond and Allinson.32 Hammond and physical space may break down. The spatial Allinson contend that a metaphor need not be metaphor may be useful as a pedagogical tool accurate at all levels of description to be useful: to introduce the novice user to the system and to aid in the initial organization of his or her Furthermore, a metaphor which is self-evi- thoughts about this new environment, but may dently inappropriate at some levels of descrip- also impede the user from developing a more tion will be more successful than one which is in-depth and explicit conceptualization of the ambiguous: no attempts will be made to invoke hypertext environment. As Pylyshyn notes,4 mappings at the inappropriate level. (p. 83)

One must distinguish between the general An example of a metaphor mismatch is ex- programmatic enterprise of trying to illumi- hibited in the metaphor of word processor as nate a new phenomenon and the much more typewriter. On a typewriter, the space bar demanding goal of establishing the validity of moves the typing position, whereas on a com- an explanatory theoretical principle. (p. 548) puter keyboard the space bar inserts a space. Novice users are often confused when the Therefore, one goal in metaphor use within spacebar does not produce the same result on psychology is to define and explain the rela- a computer that it does on a typewriter.32 tions between the source and the target domain Although mismatches are expected, these re- in the metaphor, and the cognitive processes searchers also advocate a more systematic as- involved in executing the task associated with sessment of metaphor to interface design. Car- that metaphor. It seems the spatial metaphor roll et al.38 argued that three aspects of analysis 30 BOECHLER must be undertaken to properly assess the value spatial properties and how do these properties of an interface metaphor: operational, struc- map onto “hyperspace?” tural, and pragmatic analyses. Operational analysis entails the measurement of behavioral Metric models of space effects (e.g., improvements in performance and learning). Structural analysis involves making Real world, everyday, physical space is gen- explicit the mappings of referent domain to tar- erally described in terms of geometric proper- get domains, a description of primitives and re- ties, specifically Euclidean or metric properties. lations between primitives in the referent and Three-dimensional space is denoted by the geo- target domains. Structural analysis is necessary metric relationships between three points, X, Y, to determine the scope of the metaphor. Prag- and Z. These relationships include such prop- matic analysis examines the practical context of erties as direction and distance. Metric space the metaphor. This includes examining the adheres to three geometric axioms: minimality, user’s goal associated with using the metaphor symmetry, and triangle inequality.39 These ax- as well as the impact on the user of the flaws in ioms will be described in detail presently. the metaphor (e.g., mismatches). This is similar Stanton and Baber35 express concern that this to Holyoak’s3 focus on similarity, structure, and everyday definition of space has been inap- purpose in a psychological approach to meta- propriately applied to hyperspace (which they phor validation. Carroll et al.38 emphasized that refer to as “electronic space”). They state: all three types of analyses are necessary to fully understand the dynamics between metaphor . . . the basic concept of hypertext in computer and user and, hence, to assess the overall use- science terms, is an n-dimensional space which fulness of the metaphor. can be traversed by moving through links. ‘Space’ as used in this context has a well defined meaning, as the collection of objects and An adequate analysis of interface metaphors activities contained within a specific domain, needs to rest on an empirical task analysis of similar to its use in finite state architecture. what users actually do, not an abstract nor- However, much current research in hypertext mative analysis of what they might do. (p. 73) appears to use the term ‘space’ in its everyday sense, that is, as a physical relationship be- Psychological and HCI researchers espouse tween objects. (pp. 235–236) similar types of metaphor assessment but toward a different focus; respectively, under- In the case of hypertext, it appears that phys- standing underlying information processing ical distance is being equated to some degree mechanisms versus usability. Both the psycho- with psychological distance. What exactly is logical and HCI approaches seem plausible for psychological distance? In hyperspace, one the evaluation of a broad metaphor like the spa- way of conceptualizing psychological distance tial metaphor. Nonetheless, in a survey of the would be an estimate of the degree of seman- literature, it seems neither approach has been tic relatedness or similarity between two win- used in a systematic way to closely examine the dows. hypertext/real-space comparison. The map- Geometric models, like the Euclidean model pings between hypertext features and real- of physical space, are often used in the psy- world features have not been empirically spec- chological literature to describe psychological ified nor have particular cognitive processes processes. Beginning in the early 1960s, many involved in the use of hypertext (e.g., degree of psychological researchers40–43 asserted that the spatial processing) been identified in detail. similarity of objects could be modeled using geometric properties. More explicitly, the sim- RELATING SPACE TO HYPERTEXT ilarity of two objects could be represented as points in some coordinate space, and that a While investigating the use of a spatial metric distance function could be used to com- metaphor, it is also important to consider def- pute the dissimilarity between two points. A initions of space itself. For example, what are metric distance function assigns a distance to HOW SPATIAL IS HYPERSPACE? 31 each pair of points based on adherence to the lenges the psychological validity of the trian- same three axioms that define physical space: gle inequality assumption . He makes compar- minimality, symmetry, and triangle inequality. isons between the following pairs of countries, The minimality axiom implies that the similar- Jamaica–Cuba, Cuba–Russia, and Jamaica– ity between an object and itself is the same for Russia. The dimension that defines similarity all objects. The assumption of the symmetry ax- between Jamaica and Cuba (geographic simi- iom is that the distance between two points is larity) is not the same as the dimension that de- the same no matter the order; that is, the dis- fines the similarity between Cuba and Russia tance between A and B is the same as the dis- (political similarity, at the time). Therefore, as- tance between B and A. The triangle inequal- signing a degree of similarity between Jamaica ity axiom assumes that the shortest distance and Russia is uninformative, as comparisons between two points is a straight line. That is, if are not made on dimensions that are valid be- A, B, and C are any three points in space, then tween the two countries. the sum of any two distances (e.g., AB, BC, AC) How well do geometric properties map onto is greater than or equal to the third distance. hyperspace? In hyperspace, distance could be Applied to similarity theory, it implies that if thought of as the number of steps between win- A is similar to B, and B is similar to C, then A dows (e.g., the number of mouse clicks), the and C cannot be very dissimilar from each number of layers to traverse in a hierarchical other.44 system, or the amount of time it takes to access Contrary to the above view, Tversky44 and one window from another.33 However, it could Tversky and Gati45 argued that although simi- be said that these variables are only epiphe- larity relations may be represented using geo- nomena of the underlying semantic relation- metric models including the axioms presented ships of the material as perceived by the au- above, these representations should not be thor, the person who has created the structure viewed as psychological theory. Tversky con- of the hypertext document. The links between tended that the minimality axiom is not ful- windows are constructed by the author based filled by some types of similarity measures. For on his or her perceptions of the relatedness of example, the probability of judging two iden- meaning between the content of each window. tical stimuli as the “same” rather than “differ- In this sense, the links are somewhat arbitrary ent” is not constant for all stimuli. In recogni- in that the degree of relatedness perceived by tion experiments of simple objects, sometimes the author may be very different from the de- an object is identified as another object more gree of relatedness perceived by the user. Also, frequently than it is identified as itself. Ac- depending on the order in which the user has cording to Tversky, the symmetry axiom does accessed certain windows, these variables may not hold for similarity relations either. Similar- not consistently represent, from user to user, ity judgments can be regarded as extensions of the underlying semantic relationships between similarity statements; that is, statements of the windows. form “a is like b.” In such a statement, a sub- The question therefore becomes, in the hy- ject and a referent are evident. The statement pertext context, how does meaning map onto is directional as it begins with the subject and space? One implication of the spatial metaphor ends with the referent. To reverse the statement is that the degree of semantic similarity be- is to lose its meaning; “b is like a” is not the tween windows can be psychologically repre- same as “a is like b.” sented as a distance. Windows that are similar As for the triangle inequality axiom, Tversky in meaning are “close together” in the “infor- stated that the implication mentioned above (if mation space,” windows that are not are “far” A is similar to B and B is similar to C, then A apart. This, in turn, implies that semantic sim- and C cannot be very dissimilar from each ilarity can be described using geometric prop- other) sets limits on the similarity between A erties, just like Euclidean geometric properties and C, as the similarity between A and B, and define physical space, hence, connections be- B and C constrains the similarity between A tween windows can be conceptualized as and C. Tversky provided an example that chal- “space.” Although a “metric” distance is not 32 BOECHLER applied to each pair of hypertext windows, be- into a set of basic primitives. Primitives can be cause of the comparison to real space prompted thought of as both elements and relationships by the spatial metaphor, there seems to be an that are necessary to convey the fundamental underlying assumption that some sort of dis- attributes of a phenomenon. Golledge46 identi- tance relations exist. fies and describes four spatial primitives: (1) However, as outlined above, Tversky and identity, (2) location, (3) magnitude, and (4) Gati44,45 provide some compelling reasons for time. Golledge describes providing identity “as why psychological estimates of similarity the process of equating an occurrence with a should not be represented geometrically (that name or label. The purpose behind this differ- is, as a metric distance) for all stimuli in all sit- entiation is to allow occurrences to be recog- uations. The hyperspace environment seems to nized and their uniqueness evaluated” (p. 104). violate all the axioms necessary for a geomet- As Golledge notes, a problem with identity is ric similarity model. The minimality axiom is that, at times, two labels may be given to the not always accurately adhered to with simple same entity (e.g., a street is known by different stimuli. Referring to semantic similarity be- names to different people). In hypertext, if an tween hypertext windows, this problem is in- identity is not universally acknowledged, not tensified due to the complexity of each window only may different users be unable to recognize and the sheer number of windows being ac- an item as the same item but also individual cessed in large documents. The symmetry ax- users may confuse different items. iom is also violated. Once a new window is ac- The second primitive, location, is denoted as cessed, the information the user encodes there “information about where an occurrence exists goes with him or her back to the previous win- within the totality of the environment” (p.104). dow, where that window is likely to be viewed The way location is specified can be different in a slightly different manner. Therefore, depending on the environment being consid- knowledge-wise, the initial “distance” from the ered. The spatial terms and reference points first window to the second or subsequent win- that describe an entity’s location will be differ- dows is not the same as the “distance” from the ent if that entity is a continent rather than a second window back to the first. The “dis- country, a city rather than a building or an ob- tance” has been altered merely through the ac- ject. Locational information includes informa- cessing of the second window and its contents. tion that describes relations between an entity As for the triangle inequality axiom, different and other entities within that specific environ- windows may or may not share the dimensions ment. Understanding the idea of location in hy- that other pairs of windows share just as, in pertext has several obstacles. How do we de- Tversky’s example, Jamaica, Cuba, and Russia scribe different categories of environments do not share the same dimensions. (e.g., what is global, what is local)? How do we This discussion of similarity theory has been specify location within that environment? An applied toward the validity of using a geomet- item in hyperspace has no absolute location or ric model for semantic similarity in hypertext even a consistent relative location. just as a geometric model is used to describe From the property of location, several de- real space (Euclidean properties). The depth of rived concepts emerge: (1) distance, (2) direc- comparison in this example may be excessive, tion, (3) sequence and order, and (4) connec- but it is a demonstration of how unspecified tion and linkage. Golledge46 defines distance as the parameters of the spatial metaphor remain. the “interval between the locations of occur- We are still left with an ill-defined and vague rences” (p.106). To quantify distance, location sense of how space is related to the hypertext must be clearly specified. Direction can only be environment and the elements within it. specified within an established frame of reference. Sequence and order are dependent on definitions of distance and direction as well. Con- Spatial primitives nection and linkage are related to principles of A less formal way of defining space is proximity and similarity which, again, are in- through decomposition of its characteristics extricably tied to the specification of location. HOW SPATIAL IS HYPERSPACE? 33

The relation between hyperspace and these de- behaviors in physical space should be possible rived concepts cannot be determined until the on several levels: Do people hold spatial rep- relation between hyperspace and location is resentations in memory of the hypertext envi- clearly understood. ronment? Do novice hypertext users go The third primitive, magnitude, refers to the through the same process that people in novel size of an identity. In instances where we can physical environments do? Do children show define “how much” with a single number, the similar patterns of development in hypertext measurement of magnitude is straightforward navigation as real world navigation? To ad- (e.g., how many cows are in a field). In other dress these questions we must have a brief look instances, Golledge concedes that magnitude is at spatial cognition and how it develops in real- difficult to measure. For example, do we define world environments. a city by the area covered, the size of the pop- The study of spatial cognition is character- ulation, or by the boundaries of specific activ- ized by diverse approaches in the choice of set- ities? With hypertext, do we measure magni- tings and paradigms. For example, subjects tude by the number of nodes in a document, have been asked to find their way,47 to locate the degree of linkage within the document, or an object in real space,48 as well as to imagine the degree of linkage to other documents how an object might look if it was rotated to a within a network? It is not clear how magni- different orientation.49 Small-scale environ- tude may be defined in hyperspace. ments such as a single room where all contents The fourth primitive, time, denotes the per- are visible have been studied50 and, conversely, manency of an occurrence. Does an occurrence large environments, such as the layout of a exist at a later point in time just as it exists in campus or city, where all locations are not this moment? Does an occurrence retain its lo- within view.51 Researchers have been inter- cation over time? The relation between time ested in people’s knowledge of particular and location is particularly tenuous in hyper- places as well as their knowledge of abstract text as we may not even be able to denote lo- spatial concepts. Subsequently, there are a va- cation itself. As well, large systems of hyper- riety of definitions of spatial cognition, each text documents, such as the Web, are dynamic. prompted by and constrained by the environ- Using a particular site as a landmark for locat- ment under investigation. ing other sites will be ineffective as sites can be In general terms, spatial cognition is the removed, linking can be changed, and the gen- knowledge and mental representation of the eral appearance of the site may be altered. Be- structure of space and the location and relation yond the issue of location, unlike real space, of objects or entities within that space, along navigation time in hypertext is not correlated with the knowledge of when and how to use to “distance.” Changing locations between this stored information and the ability to think windows that are unrelated can be achieved about the mental representations themselves.52 just as quickly as moving between windows This definition includes two parts: the infor- that are closely related. mational system itself and the individual’s abil- Whether we approach the hypertext/real ity to access and make use of that information. space comparison from the axioms of Euclid- The informational system has been referred to ean space or a list of spatial primitives, identi- as a cognitive map, a term first used by Tol- fying discrepancies between the two environ- man.53 In this paper, the term cognitive map ments will inform us about the value of the will be used to refer to the informational sys- spatial metaphor toward enhanced informa- tem that is part of and supports spatial cognition access. tion.

Cognitive maps. Cognitive maps are long- Spatial cognition term memory representations of spatial infor- If spatial cognition plays an important role mation. As we cannot view cognitive maps di- in how people interact with hypertext, com- rectly, we must infer their features from parisons between behaviors in hyperspace and behavior during specific tasks. Consequently, 34 BOECHLER empirical evidence involves these representa- time. According to Kitchin, this is an important tions as they are applied to the performance of aspect of the cognitive map, that it is con- a particular action. External forms produced by structed to combine the concepts of place and behavior (e.g., sketch maps, verbal protocols, action. Cognitive maps are individual con- quantitative and structural analyses), are often structions for individual events: there is no sin- called spatial products, a term originally used gular cognitive map that retains all the spatial by Liben.52 Spatial products give us some in- information and knowledge we’ve acquired. dication of the nature of underlying spatial rep- They are dynamic; that is, our knowledge (cog- resentations. Behavioral evidence suggests nitive map) influences the experiences we have what the rules and operations are that govern and our experiences alter our knowledge (cog- the use of spatial representations.54 nitive map). The term cognitive map often elicits the as- More recent research into the nature of cog- sumption that this representation is an image- nitive maps has suggested that their hierar- based, map-like mental representation. The ac- chical structure is important. Hierarchies are tual form of cognitive maps has been highly organized by semantic and conceptual related- debated. Some researchers have suggested that ness, not by spatial relations. Several studies the cognitive map is analogous to a carto- provide evidence that cognitive maps have a graphic map, a Euclidean model of the world.55 hierarchical structure; that is, entities are en- This is not to say that cognitive maps are exact coded at different levels or regions (e.g., par- replications of features on a map. They have ticular objects in relation to sets of objects) been described as “schematic, sketchy, incom- based on their conceptual similarity or dissim- plete, distorted, and otherwise simplified and ilarity rather than their actual spatial location. idiosyncratic.” 56 For example, Stevens and Coupe59 found Other researchers have suggested a concep- that people are influenced by superordinate tual-propositional, rather than image-based, spatial relations of geographic locations. They theory of spatial information encoding.57 In this asked subjects to judge the spatial relations of view, both verbal and visual information is different cities (e.g., Is Montreal north or south stored as abstract conceptual propositions. of Seattle?). People often misjudged the loca- Concepts are embedded in a propositional net- tion of different cities based on the state or work and conceptual meanings are partially country the city is situated in (e.g., Montreal is dependent on their relations to other concepts in Canada and Canada is north of the United in the network. If concepts are close together in States, therefore Seattle must be south of Mon- the propositional network, they will function treal). The superordinate category of country as cues for the retrieval of one another. Alter- influenced judgements about the subordinate natively, some researchers57,58 have argued category of cities, irrespective of their spatial that the term “map” should only be thought of locations. as a hypothetical construct for the spatial sys- McNamara et al.60 submitted subjects’ recall tem that we accept as existing and influencing protocols of a spatial layout to an algorithm that our spatial decisions. constructs hierarchical trees consistent with the Aside from the debate about the form of cog- internal organization of the protocols. They nitive maps, Kitchin58 summarizes the types of found that subjects’ distance estimations and information that cognitive maps contain. Cog- spatial priming was dependent on whether pairs nitive maps are selective, subjective represen- of objects appeared in the same subtree or dif- tations of the environment that preserve the lo- ferent subtrees of the hierarchy rather than each cational and attributional features, including object’s spatial location in the total array. subjective emotional responses to those attrib- Both of the above studies highlight the im- utes, that are most informative for the individ- portance of categorization and meaning in the ual perceiving that environment. The attributes organization of mental representations of spa- that designate the structure of the map (land- tial information. If hierarchical structure is im- marks or labels) are markers for action that portant to the retrieval of spatial memories of should take place at that point in space and objects in physical space, it is plausible that hi- HOW SPATIAL IS HYPERSPACE? 35 erarchical structure is even more relevant than marks are coordinated in a sequence, which re- spatial position to the recall of objects in hy- sults in path or route representations. This perspace, where objects have no absolute posi- route knowledge is comprised of both spatial tion or even consistent relative position. and temporal relations between landmarks. As route representations become more refined, a The development of spatial cognition representation of the entire space forms, a survey representation that includes the relations Much of the research on early spatial cogni- between landmarks and routes. Thus, with in- tive development focused on the development creased familiarity of the environment, spatial of children’s cognition related to small spaces, representations gradually incorporate more so- particularly, the analysis of the children’s phisticated versions of landmark, then route, changing frames of reference. As they develop, then survey knowledge; survey knowledge be- children switch from egocentric frames of ref- ing qualitatively different from the first two be- erence to nonegocentric or geographic frames cause it incorporates the understanding of met- 61 of reference. Egocentric references are spatial ric distance relations between landmarks and indicators pertinent to the child’s own body routes. Siegel and White’s view remained the and point of view (e.g., to the left or right of dominant framework throughout the 1970s me). Geographic references are spatial indica- and 1980s, and has been cited in the hypertext tors that are tied to the space as a whole (e.g., literature.15,30,64 the relation between two objects in a room or More recently, Montello65 argued that the fi- the walls of a room). Children typically switch nal acquisition of survey knowledge is a quan- their frame of reference from egocentric to ge- titative rather than a qualitative change. He as- ographic references between 6 and 16 months serts that metric knowledge is being encoded of age. Beyond 16 months, changes in the right from the earliest learning stages, and that choices of different geographic reference sys- survey knowledge appears as the qualitative tems develop; that is, as they get older children outcome of the integration of spatial knowl- increasingly choose more and more distal ver- edge, including metric knowledge, about sep- sus proximal systems (e.g., the walls of a room arately learned places. Montello describes spa- vs. the furniture in a room). As children’s tial knowledge as consisting of both nonmetric frames of reference radiate further away from and metric information beginning with the first their person, and include object–object rela- exposure to a new place. He also asserts that tions as well as body–object relations, they are some types of spatial information are stored in increasingly able to deal with larger environ- a linguistic format and comes to the conclusion 62 ments. By 12 years of age, children’s compe- that “there exist two or more distinct subsys- tence in real-world large-space navigation tems that represent spatial information in dif- 51 matches that of adults. ferent formats” (p. 17). Montello cites several authors who present converging evidence and Spatial knowledge acquisition in adults. In a arrive at comparable conclusions. For example, highly influential article, Siegel and White63 McNamara et al.66 referred to three formats of suggest that the ability to act efficiently within spatial knowledge—temporal strings, proposi- large-scale environments is based on the men- tions, and metric spatial representations. tal integration of multiple views. Because large Kuipers67 proposed that spatial knowledge is environments cannot be viewed all at once, stored in many disconnected components that some type of mental–spatial representation include separate metric and topological com- must be constructed to comprehend the entire ponents. Kosslyn68 suggested that the right space. Siegel and White proposed that the ac- hemisphere of the brain is specialized for pro- quisition of these spatial representations fol- cessing coordinate spatial representations lows a distinctive course, as a series of stages, while the left hemisphere is specialized for pro- landmark recognition being the first in which cessing categorical representations. individuals learn to discern and remember sep- Three questions emerge from Montello’s po- arate landmarks. In the second stage, land- sition: (1) If metric knowledge is so pervasive 36 BOECHLER throughout the acquisition of knowledge of the surrounding environment provides addi- large spaces, how relevant is real world navigational retrieval cues for that action. In hyper- tion to hypertext as metric distance may not be text, we view only one window at a time. If the a property of hypertext?; (2) Do spatial aids window itself is not a strong enough cue to re- prompt the user to develop some form of men- trieve the action from memory (e.g., select a tal representation that includes metric distances particular icon to another window) and there between elements?; (3) Integration of multiple are no navigational aids to help orient the user, types of spatial information is cognitively de- the user must make a random selection of manding69; therefore, what is gained by adding icons. components to hypertext (e.g., spatial maps) that In contrast, real-world environments have an may force another knowledge subsystem (in this abundance of perceptual information; that is, case metric knowledge) to be engaged? they provide a rich collection of features and An alternative perspective is that navigation attributes on which we can base spatial deci- may be route-based, that is, route knowledge sions. Perceptual information allows us to is not a prerequisite for or subset of survey make predictions about the future positions of knowledge but rather route and survey knowl- objects in the environment as we move through edge are acquired and represented separately. space. For example, in a city setting, distal land- Hirtle and Hudson70 found that subjects could marks such as tall buildings are occluded by acquire route knowledge from both slide pre- closer buildings. As we move through space, sentations and maps but could only acquire the degree of occlusion changes, giving us clues survey knowledge from maps. This suggests as to the future position of the buildings that that, with hypertext, even without global sur- are not entirely in our view in that moment. vey knowledge of a document, users should be Hypertext environments are featurally im- able to encode sequences of landmarks into poverished compared to real-world environ- routes and integrate separate routes for a local ments. This may make the attributes that are representation of parts of the document. How- present more salient, but it also provides fewer ever, even with a route-based explanation of opportunities for the user to establish multiple hypertext use, we still encounter the problem reference points that designate the user’s posi- of distance definition. A route is defined by the tion along a route. There is no perceptual in- sequence of and distances between landmarks. formation that is inherent to the hyperspace en- In hypertext, distance is not easily measured vironment that helps us make predictions and is inconsistent from one traversal of the about the future positions of objects. document to the next. The importance of children developing broader frames of reference outside of their own bodies was mentioned earlier in the sec- A PRELIMINARY COMPARISON OF tion on spatial cognition. The importance of us- HYPERSPACE AND PHYSICAL ing the body as a reference for spatial decision- 71 SPACE FEATURES making has been argued for adults as well. Body position as a perceptual cue and the Metric distance may be one property that is sense-of-self within the environment as a phys- not shared by hyperspace and physical space. ical cue are not characteristics of hyperspace. In the following section several other possible disparities are suggested as examples of the The nature of landmarks types of issues that need to be investigated to evaluate the appropriateness of the spatial In real space, landmarks have no inherent metaphor. meaning. Their meaning is defined by the action attached to them. Landmarks have a sequential relationship but not a semantic rela- Perceptual cues tionship. For example, a tree has no shared As mentioned previously, landmarks are im- meaning with a car unless an action pairs the portant in navigation because they are a cue for two together such as, “I must walk to the car a specific action to be carried out. In real space, and turn left toward the tree.” In hypertext, HOW SPATIAL IS HYPERSPACE? 37 landmarks, such as individual windows, have mental space equivalent to the size of a neigh- their own semantic content that is related to borhood or a city or province? In the real world, other landmarks aside from their possible se- we have different cognitive strategies for deal- quential or spatial relationship. ing with different sizes of spatial configura- A sequence of landmarks makes up a route. tions. How well hypertext cognitively maps In real space, the length of that route and the onto physical space may be dependent on ap- spaces between landmarks can be defined by a propriate scale matches. distance. Distance in real space is correlated with time, that is, the longer the distance the more time it takes to traverse it. In hyperspace, HYPERTEXT NAVIGATION this correlation does not hold, partly because we have no way of accurately measuring dis- To this point, this paper has focused on a tance but also because hypertext is dynamic. broad overview of some of the issues regard- Distances do not remain static but change de- ing the spatial metaphor. The following section pending on the order of access the individual presents specific studies focused on the contri- uses at any given time or the different prefer- bution of spatial navigational aids to hypertext ences expressed by different users. performance. Information access in hypertext documents Spatial mental representation is impacted by the structure of the document itself outside of the navigational aids that may Survey knowledge in a large hypertext doc- be provided.72 There are three levels at which ument can only be achieved through naviga- this may occur: (1) The inherent structure of the tional aids such as maps and content lists. This material (Is hierarchical organization an intrin- means that it is possible that survey knowledge sic quality of the material regardless of pre- is mentally represented in a different form than sentation?); (2) The organization of the mater- route knowledge where the user has visually ial within the document or document structure experienced the pages. If survey knowledge is cues (Are pages from subordinate categories encoded in a different form than route knowl- differentiated from pages in superordinate cat- edge, it must undergo some kind of mental ma- egories?); and (3) The linking structure (Does it nipulation so as to track it to route knowledge. reflect both of the above or, in and of itself, does This increases the cognitive demands of the it impose structure or constraints on the docu- task. ment?). These are global features of the hyper- Contrary to this, in real world navigation dis- text document that are beyond the focus of this tal landmarks and the general lay of the land- paper. However, we need to be aware of and scape are visible. Through these cues, survey explicitly report these features during the knowledge can be partially acquired in the course of investigating navigational aids. same manner as route knowledge; it is an ex- pansion of the cognitive map developed for Measures route knowledge. The accumulation of metric and nonmetric information occurs from the How is navigational performance typically outset, that is, survey knowledge begins to be measured? Two categories of measurement en- encoded in the same manner, at the same time capsulate most of the measures typically used as route knowledge. A presentation of a survey in hypertext research: efficiency and effective- map in the homepage of a document provides ness measures.73 survey information at the outset as well, but in Efficiency measures are based on speed and a different form than route knowledge. the number of steps taken to complete an information search. Efficiency measures often include the number of nodes visited in relation A question of scale to the entire document, the number of excess If physical space is cognitively related to hy- nodes visited, the number of nodes revisited, perspace, how many nodes does a hypertext the time taken on each node of the document, document need to have to approximate the and the total task time. 38 BOECHLER

Effectiveness measures focus on the user’s is to be a truly useful informational tool both search accuracy as well as his or her recall and measures would seem important. Making de- understanding of the structure of the docu- cisions quickly and accessing the least win- ment. Effectiveness measures typically include dows necessary is relevant, but if users are not measures of recall of the elements within and able to find the information they need (incom- the structure of the document. This would in- plete or inaccurate search) or recall how they clude the number of node titles remembered, found it, the usefulness of hypertext could be number of node titles filled into a blank severely constrained. Therefore, it would seem overview, or a sketch map of the document.12 that a successful navigational aid should in- Accuracy is also an effectiveness measure de- crease both efficiency and effectiveness. noted by the number of correct answers to questions that require the user to conduct an Types of navigational aids information search. A distinction needs to be made between nav- A spatial map, spatial overview, or graphi- igation effectiveness and learning of the docu- cal browser is a visual representation of the ment material. In contrast to effectiveness, structure of the document that depicts the hy- learning involves a deeper understanding of pertext nodes and the links that connect the content of the document and is often mea- them.8,9,74 These are usually in a diagrammatic sured by the study or review time per node, form such as block diagrams, diagrams orga- the number of main ideas or concepts re- nized around a central term (spider map), or tained,74 and the number of concepts retained hierarchically ordered tree diagrams (Figs. 1 over time.9 and 2). It is meaningful to clarify the distinctions be- Nonspatial navigational aids or textual aids tween efficiency measures, effectiveness mea- include alphabetical or arbitrarily ordered con- sures, and enduring learning as the first two tent lists, conceptually ordered content lists, are very important to information access in hy- and indices.75 It is difficult to pinpoint what pertext, while the third measure is pertinent to features distinguish a textual from a spatial educational value. aid. I suggest that lists that assign a spatial lo- Several studies that have found positive ef- cation through features such as indentation76 fects for spatially based navigational aids have should be thought of as a spatial aid as they reported an increase in efficiency measures but provide a spatial location for each term in the not in effectiveness measures.8,9,14 If hypertext layout.

FIG. 1. An example of a spatially ordered navigation aid. HOW SPATIAL IS HYPERSPACE? 39

FIG. 2. An example of a spatial-hierarchically ordered navigation aid.

The effects of navigational aids detailed comparisons of the specific features of each type of navigation tool. It is still not clear Most hypertext tasks can be described within which specific features may contribute to this the broad classification of open or closed effect. tasks.77 Open tasks have a general goal of browsing or exploring the document. Closed Effects of navigational aids on closed tasks tasks have a specific goal of finding a particular piece of information. The effects of naviga- McDonald and Stevenson9 reported on two tional tools on both types of tasks have been studies examining the effects of navigational studied. The following section describes first a aids on navigation and learning. In the first general metaanalysis on navigational aids, task study, the effect of a localized spatial map, a types, and user characteristics, followed by textual content list, and no aid were compared. specific examples of studies on closed and open The map consisted of labels with connecting tasks. lines indicating the links between nodes. Not Chen and Rada73 conducted a metaanalysis all nodes were linked. The content list was a on 20 experimental studies and 3 doctoral dis- scrollable list of all the nodes in the document. sertations that appeared between 1988 and The navigational aids were not interactive in 1993. Generally, the metaanalysis indicated that users could not access pages via the aids that the effects of navigational tools such as in- but had to navigate through the pages indi- dices, table of contents, and graphical maps in- cated by the map or list. The aids were dis- teract strongly with the type of task involved played in separate windows from the text con- (open or closed). tent of the document. Subjects were asked to More specifically, for effectiveness scores first find 10 target nodes and immediately af- (based on various measures of accuracy), task ter were asked 20 questions regarding infor- complexity showed the greatest effect size, fol- mation in the text. The navigation measures lowed by graphical maps. For efficiency scores were: the mean time taken to access the target (based on various speed measures), task com- nodes and the number of additional nodes plexity and spatial ability showed the largest opened beyond that determined to be the op- effects. Also, spatial ability interacted with nav- timal route. The learning measures were: the igational aid on measures of efficiency (speed). number of questions answered correctly and These results suggest that spatial processing the mean number of node titles recalled during may play a role in hypertext processing. How- free recall. The results show that subjects in the ever, as a metaanalysis is an analysis of com- map condition outperformed both the content mon features across studies, it does not make list and no aid conditions on both measures of 40 BOECHLER navigation efficiency. However, no significant page within the document. Performance was differences were found between the three con- measured on the accuracy of the information ditions on accuracy on the test questions. The search (answers to the test questions), time number of titles recalled were comparable be- spent on each screen page, frequency of visits tween the spatial map and the contents list but to each information node, frequency of visits to were significantly less for the no aid condition. the map, and time spent on the navigational The spatial map did not produce better recall map. Performance on these measures was com- than the content list. pared to a predefined, optimal path for that In McDonald and Stevenson’s second study, particular test item as determined by the au- a spatial map and conceptual map were com- thors. Subjects were divided into three groups pared to a no aid condition. The conceptual according to map access (both frequency and map was similar to the spatial map described time spent on the map). Subjects who visited for the first study but with the added feature the map frequently and/or spent more time on of link descriptions. The same measures were the map accessed more unnecessary nodes than recorded. Subjects again found 10 target nodes other users. The authors conclude that the “fre- but then answered 40 questions based on the quency of visiting the map and time spent on text, 20 factual questions and 20 questions it do not guarantee orientation to the process meant to tap into deeper conceptual learning. of organizing and acquiring relevant informa- This session was referred to as the acquisition tion” (p. 184). phase. A week later, subjects returned to an- In another search task study, Leventhal et swer 40 additional questions to test for long- al.15 found that measures of navigational per- term retention (the retention phase). As in the formance were not correlated with use of a hi- first study, significant differences occurred be- erarchical overview. Correlations between ac- tween all conditions on both efficiency mea- curacy and speed measures and both the sures. On the effectiveness measures, there was percentage and total number of visits to two no difference between the spatial and concep- types of hierarchical overview cards were not tual map conditions on the accuracy of factual significant. questions in the acquisition phase. Both were Stanton et al.12 found evidence against facil- better than the no aid condition. In the reten- itation of navigation maps on effectiveness tion phase, conceptual map subjects answered measures. Subjects were divided into map and more factual questions correctly than the spa- no-map conditions and were asked to search tial map and no aid condition. Regarding the out the answers for a sentence completion task. conceptual questions, in the acquisition phase, They were then asked to draw an outline of the subjects in the conceptual map and no aid con- hypertext document on paper. The results in- ditions answered more questions correctly than dicated that the no-map subjects outperformed the spatial map subjects. In the retention phase, the map subjects on accuracy of the sentence there was a difference between all three groups; completion task. Subjects in the no-map condi- conceptual map subjects answering correctly tion were also better able to produce a draw- most often, followed by the no aid group, and ing of the document structure and reported finally the spatial map group. These results feeling more in control of the search situation. suggest that a conceptual map is important for learning, whereas a spatial map increases effi- Effects of navigational aids on open tasks ciency but not effectiveness. Dias and Sousa13 studied the effects on per- Dee-Lucas and Larkin74 investigated the ef- formance of a navigational map, described as fects of structured (a map) versus unstructured a content index that provides a global overview (an alphabetical list) overviews on recall. The of the document as well as a representation of map consisted of a hierarchical diagram with the path taken by the user. Subjects were asked one central heading and three levels of sub- to find the answers to 12 questions. A map icon headings. Through the placement of the head- allowed the users to access the overview. From ings, referred to as unit titles, the map indicated the overview, the subjects could access any the subordinate and superordinate relations HOW SPATIAL IS HYPERSPACE? 41 between the unit titles. The list provided no in- tween the browser and no-browser groups. The dication of the relations between the unit titles. only significant differences occurred in effi- Both overviews were interactive, in that as the ciency measures: the number of nodes accessed reader finished a unit he or she was required and the number of nodes revisited. Subjects in to return to the overview to select and link up the browser groups accessed a higher propor- to a new unit title. Both document formats con- tion of the total available nodes and exhibited tained nine units. Participants were asked to less revisitation than the no-browser subjects. read all material first in a specified order and Subsequently, the authors suggested that pre- then were free to review any material they sentation of a graphical browser enhanced the chose. Participants were then tested on both re- efficiency of browsing but did not improve call and study strategy measures. The recall users’ performance on measures of effective- measures included the number of unit titles re- ness. called, the proportion of text propositions An important aspect to note about a number (main ideas) recalled, the proportion of propo- of the studies comparing spatial navigation sitions recalled per unit, and the proportion of tools versus content lists is that hierarchical or- units with at least one proposition recalled. On ganization and spatial location are often con- these measures of recall there were no differ- founded.9,10,74 By placing node titles within a ences between the map and the list groups. hierarchical framework, nodes are automati- However, the map group were more able to re- cally assigned a spatial location. Conversely, call the title locations than the list group, and providing a basic content list removes not only the list group spent more time on the overview. spatial location but also the hierarchical infor- Wenger and Payne14 investigated whether mation that highlights the conceptual relations the presence of a graphical browser impacted between node titles. Although any time one la- the comprehension and retention of the content bel is visually differentiated from another some and structure of a hypertext document. In this sort of spatial context is created, this effect study, the graphical browser indicated the might be reduced by using typographical cues structure of the document in a hierarchical (e.g., colored or bolded font) on simple content fashion and also, through the highlighting of list labels to denote which topics are nested visited nodes, allowed the user to see which within other topics (providing information portions of the document had been accessed. about conceptual relations) while keeping at a To minimize imposing a certain order of view- minimum the assignment of an explicit spatial ing, users were allowed to access the document location. However, there are drawbacks to this at any entry point by choosing, on the first method as well. The hierarchy of the document node, from an unstructured alphabetical listing could be quite broad (have many labels at the of the contents. Both the browser group and the same level of the hierarchy), but could not be no-browser group had this option. The browser very deep (have many levels within the hier- group could return to the browser from any archy) or the list itself would become overly node in the document. The subjects were in- complex. structed to read through all the nodes and indicate when they felt they had accessed all available nodes. At the end of the text, in a THE ROLE OF timed recall task, users were required to note COGNITIVE PSYCHOLOGY as many as the node titles as they could remember. They then completed multiple choice The previous section provided examples of tests as an indicator of their comprehension of variables that can contribute to navigation out- the text. Finally, to tap into the users’ knowl- comes. Task type, navigational aids, and over- edge of the text structure, users were presented all document structure all impact information with pairs of node titles and asked to state search. The section also illustrates through em- whether the nodes were linked within the pre- pirical studies the inconclusiveness of data to- vious text. On all three of these effectiveness ward answering the central questions raised in measures, results showed no differences be- this article. 42 BOECHLER

The following section is an outline of some of processing. Particular to hypertext, if sub- ways in which traditional cognitive psychology jects are asked to engage in a hypertext search paradigms could be used to investigate some task and this involves spatial processing, then of the issues described throughout this paper. a second task also involving spatial processing should degrade performance on the search task Spatial mental representations: Accessing outcomes. Conversely, a second task involving underlying mental representations and matching some other type of processing (e.g., auditory to the conceptual properties of the material processing) should not impact performance to such a degree. If information in hypertext is en- Do spatially based aids facilitate the con- coded spatially, the spatial task will interfere struction of a mental representation of the doc- with the encoding of the document structure ument that reflects the organization of the ma- and affect measures such as the recall of node terial? The same methodologies used to titles. The spatial task could involve monitor- examine spatial cognition in physical space ing the appearance and location of a visual ob- could be applied to examine cognitions in hy- ject on screen. The nonspatial task could in- perspace. For example, McNamara et al.’s60 hi- volve distinguishing several auditory tones erarchical tree analysis for assessing the un- from one another. derlying spatial representations for physical spatial layouts could be applied to users’ mental representations of hypertext documents. Metaphor validation: Mapping features of Subjects’ recall protocols of the hypertext doc- physical space to features of hyperspace ument structure could be subjected to McNa- Is the spatial metaphor appropriate for hy- mara et al.’s algorithm, which constructs hier- pertext? It is prudent to take a closer look at archical trees consistent with the internal the benefits and limitations of this all-encom- organization of the protocols. Subjects’ distance passing spatial metaphor from the same per- estimations and spatial priming could be used spectives (operational, structural, and prag- to determine whether pairs of windows that are matic) and with the same rigor that are used to conceptually related appeared in the same sub- evaluate metaphors of a narrower focus. Iden- tree or different subtrees of the hierarchy, tifying the mismatches is important. Even if rather than in the order of access or some other users may not be able to report them, mis- type of spatial organization. When comparing matches may provide implicit knowledge to navigation aids with different degrees of spa- users that may negatively affect their mental tial information, this method could reveal representations of the hypertext environment. which is most important to the user, concep- Initially, a structural analysis could be un- tual or spatial information. dertaken, as this is the type of metaphor analysis often used in both psychological and HCI The role of spatial processing: Identifying the research. Also, structural feature mapping be- presence of spatial processing tween hyperspace and physical space would be Is spatial processing involved in the execu- useful toward setting the parameters of the tion of a hypertext search task? The dual-task metaphor, a framework on which to pursue the interference paradigm is often used in cogni- operational and pragmatic analyses. A ques- tive psychology to detect the presence of spe- tion raised earlier may be a relevant starting cific types of cognitive processing.78 The basis point: Is metric distance a property of hyper- of this task is that two tasks executed concur- text? If metric distance is a property of hyper- rently will produce degraded performance, text then the knowledge we have about how particularly if they involve the same type of people navigate in physical space may be rele- processing. When a subject is presented with a vant to navigation in hyperspace. One way of second task that requires a different type of investigating this issue is to determine whether processing than the task already engaged in, hyperspace violates the three axioms that de- the second task will interfere less with perfor- fine physical space. For example, the symme- mance than a task that demands the same type try axiom could be evaluated through the col- HOW SPATIAL IS HYPERSPACE? 43 lection of distance ratings from subjects asked location with other windows branching off. An to make comparisons between pairs of win- alternate view is that revisitation may be a dows. Distance ratings for all pairs and pair or- symptom of confused identity. Because of ders (e.g., the distance from A to B as well as poorly differentiated labels or visual appear- the distance from B to A) could be collected and ance, a subject may revisit a window mistak- the matrix of this data could be evaluated for ing it for a new unopened window or confus- symmetry. ing it with a different, previously accessed If the data does not support a metric repre- window. Again, including qualitative data may sentation (e.g., shows asymmetric distances), be illuminating. Verbal justifications of revisi- further evaluation based on adherence to the tations could be collected during the search spatial primitives outlined by Golledge46 tasks to tap into these differences. would be revealing. Golledge is cautious about Also, verbal protocol recording can be un- the application of spatial primitives as he de- dertaken during the navigation task. Subjects fines them for several reasons. Some of these could be trained to “think aloud” their strate- definitions can only be expressed and investi- gies and thoughts as they proceed through a gated using imprecise, natural language (e.g., hypertext document. Previous research on pro- How can we quantify identity?) rather than in tocol analysis suggests the task of navigating a the unambiguous technical terms used in com- document would be conducive to the “think puter spatial systems such as Global Position- aloud” approach and produce the type of pro- ing Systems. Golledge expresses concern that tocol data that can be analyzed in detail.79 such systems entail technical language based on imprecise definitions of spatial primitives: CONCLUSION We do this without really knowing how error accumulates as spatial primitives are joined or From the above review it is apparent that as spatial derivatives are deduced from spa- there are many complex psychological factors tial primitives. We do this without fully un- involved in the use of the World Wide Web. derstanding that our fuzzy and error-ridden Because of the vast potential for the Web to im- concepts might be inappropriately transferred pact millions of people, it is important that we to the exact and precise domain of the com- develop a detailed understanding of users’ per- puter. (p. 103) ceptions of hyperspace, beginning with a thorough investigation of the most basic cognitive Therefore, examining the comparison be- processes involved, processes that have been tween hyperspace features and spatial primi- and continue to be closely examined with re- tives may be best undertaken as a combination spect to many other types of environments and of quantitative and qualitative analyses. Data stimuli. collected via tracking mechanisms would in- For the computer scientist’s goal of develop- clude quantitative variables such as the time ing optimal interfaces, information gathered spent on individual windows, the number of from such cognitive studies should aid in the times a window is revisited, and the number development of more user-friendly web sites, of excess windows opened beyond an optimal navigation tools, and search mechanisms. For path. To interpret users’ behaviors, quantita- clinical psychologists, such information may tive data such as the variables above may be highlight aspects that need to be considered in combined with data from spatial products. For the Web presentation of psychological services. example, frequent revisitations to a window This information may also shed light on the in- may indicate that a particular window is being herent aspects of hyperspace that allow for the used as a reference point or landmark. This im- development of negative behaviors associated plies that a window is being assigned a loca- with the Web. For social psychologists, under- tion relative to other windows. If a window is standing hyperspace and how it cognitively used as a landmark, sketch maps of networks maps onto physical space may predict behav- may portray that window as having a central ioral advantages, disadvantages, and solutions 44 BOECHLER related to issues about the globalization of our Maps as navigational aids in hypertext environments: society and the individual’s social place within An empirical evaluation. Journal of Educational Multi- it. For educators, such information may facili- media and Hypermedia, 1(4):431–444. 13. Dias, P., & Souza, A.P. (1997). Understanding navi- tate the presentation of instructional material gation and disorientation in hypermedia learning en- in a format that will reduce cognitive overhead vironments. Journal of Multimedia and Hypermedia, in the learner. 6(2):173–185. Understanding the cognitive processes in- 14. Wenger, M.J., & Payne, D.G. (1994). Effects of a graph- volved in hypertext use, particularly people’s ical browser on readers’ efficiency in reading hypertext. Technical Communications, 41:224–233. basic conceptualizations of hyperspace, can 15. Leventhal, L.M., Teasley, B.M., Instone, K., Rohlman, only enhance our ability to make the World D.S., & Farhat, J. (1993). Sleuthing in Hyperholmes- Wide Web an accessible information resource super™: An evaluation of using hypertext vs. a book for users of all types. Cognitive psychologists to answer questions. Behavior & Information Technol- can play an important role in the identification ogy, 12(3):149–164. and analysis of how the processes of the mind 16. Reynolds, S.B., & Dansereau, D.F. (1990). The knowledge hypermap: An alternative to hypertext. Com- interact with the Web. puters in Education, 14(5):409–416. 17. Campagnoni, F.R., & Ehrlich, K. (1990). Getting there when you don’t know where “there” is: Navigational REFERENCES strategies in a hypertext help system. SIGCHI Bulletin, 21(3):17–18. 1. Rock, I. (1983). The logic of perception. Cambridge, MA: 18. Vicentes, K.J., & Willeges, R.C. (1988). Accommodat- MIT Press. ing individual differences in searching a hierarchical 2. Goodman, N. (1978). Ways of worldmaking. Indi- file system. International Journal of Man-machine Stud- anapolis, IN: Hackett Publishing Co. ies, 29:647–668. 3. Holyoak, K.J., Thagard, P. (1996). Mental leaps: Anal- 19. Leidig, P.M. (1992). The relationship between cogni- ogy in creative thought. Cambridge, MA: MIT Press. tive styles and mental maps in hypertext assisted 4. Pylyshyn, Z.W. (1993). Metaphorical imprecision and the learning. Dissertation Abstracts International, 53(5-A): “top-down” research strategy. In: Ortony, A., (ed.) 1372. Metaphor and thought. Cambridge, MA: Cambridge 20. Bieber, M., Vitali, F., Ashman, H., Balasubramanian, University Press, pp. 420–436. V., & Oinas-Kukkonen, H. (1997). Fourth generation 5. Canter, D., Rivers, R., & Storrs, G. (1985). Character- hypermedia: Some missing links for the World Wide izing user navigation through complex data struc- Web. International Journal of Human-Computer Studies, tures. Behavior and Information Technology, 45(2):93– 47:31–65. 102. 21. Jonassen, D.H. (1986). Hypertext principles for text 6. Kim, H., & Hirtle, S.C. (1995). Spatial metaphors and and courseware design. Educational Psychologist, 214: disorientation in hypertext browsing. Behavior and In- 269–292. formation Technology, 14(4):239–250. 22. Jonassen, D.H. (1989). Hypertext/hypermedia. Engle- 7. Kim, J. (1999). An empirical study of navigational aids wood Cliffs, NJ: Educational Technology Publications. in customer interfaces. Behavior and Information Tech- 23. Schneiderman, B., & Kearsley, G. (1989). Hypertext nology, 18:213–224. hands-on: An introduction to a new way of organizing and 8. McDonald, S., & Stevenson, R. (1998). Spatial versus accessing information. NY: Addison-Wesley. conceptual maps as learning tools in hypertext. Jour- 24. Conklin, J. (1987). Hypertext: An introduction and nal of Educational Hypermedia and Multimedia, 8(1): survey. Computer, IEEE Computing Society, 20(9):17– 43–64. 41. 9. McDonald, S., & Stevenson, R. (1999). Navigation in 25. Nielsen, J. (1990). Hypertext and hypermedia. San Diego, hyperspace: An evaluation of the effects of naviga- CA: Academic Press. tional tools and subject matter expertise on browsing 26. Gygi, K. (1990). Recognizing the symptoms of hyper- and information retrieval in hypertext. Interacting with text . . . and what to do about it. In: Laurel, B., (ed.) Computers, 10:129–142. The art of human-computer interface design New York: 10. Beasley, R.E., & Waugh, M.L. (1995). Cognitive map- Addison-Wesley, Reading, pp. 279–288. ping architectures and hypermedia disorientation: An 27. Tsai, C.L. (1988). Hypertext: Technology, applications empirical study. Journal of Multimedia and Hypermedia, and research issues. Journal of Education Systems 4:239–255. 17(1):3–14. 11. Schroeder, E.E., & Grabowski, B.L. (1995). Patterns of 28. Ransom, S., Wu, X., & Schmidt, H. (1997). Disorien- exploration and learning with hypermedia. Journal of tation and cognitive overhead in hypertext systems. Educational Computing Research, 13(4):313–335. International Journal of Artificial Intelligence Tools, 12. Stanton, N.A., Taylor, R.G., & Tweedie, L.A. (1992). 6(2):227–253. HOW SPATIAL IS HYPERSPACE? 45

29. McLeese, R.A. (1989). Navigation and browsing in hy- 45. Tversky, A., & Gati, I. (1982). Similarity, separability and pertext. In: McLeese, R.A., (ed.) Hypertext: Theory into the triangle inequality. Psychological Review, 89:123–154. practice. Westport, CT: Ablex, pp. 6–44. 46. Golledge, R.G. (1995). Primitives of spatial knowl- 30. Edwards, D.W., & Hardman, L. (1989). “Lost in hy- edge. In: Nyerges, T., et al. (ed.) Cognitive aspects of hu- perspace”: Cognitive mapping navigation in a hy- man-computer interaction for geographic information sys- pertext environment. In: McLeese, R.A., (ed.) Hyper- tems. NATO/ASI Series: Kluwer, pp. 29–44. text: Theory into Practice. Westport, CT: Ablex, pp. 47. Passini, R. (1984). Wayfinding in architecture. New 90–105. York: Van Nostrum Reinhold. 31. Foss, C.L. (1989). Detecting user’s lost: Empirical stud- 48. Cornell, E.H., & Heth, C.D. (1983). Spatial cognition: ies on the browsing of hypertext. Technical Report, No. Gather studies used by preschool children. Journal of 972, INRIA, Sophia-Antipolis. Experimental Child Psychology, 35:93–110. 32. Hammond, N., & Allinson, L. (1987). The travel 49. Shepard, R.N., & Metzler, J. (1971). The mental rota- metaphor as design principle and training aid for nav- tion of three-dimensional objects. Science, 17:701–703. igating around complex systems. In: Diaper, D., 50. Acredolo, L.P. (1978). Development of spatial orien- Winder, R., (eds.) People and Computers III: Proceedings tation in infancy. Developmental Psychology, 14:224– of the third conference of the British Computer Society, Hu- 234. man-Computer Interaction Specialist Group. Cambridge: 51. Cornell, E.H., Heth, C.D., & Alberts, D. (1994). Place Cambridge University Press. recognition and wayfinding by children and adults. 33. Shum, S. (1990). Real and virtual spaces: Mapping Memory and Cognition, 22(6):633–643. from spatial cognition to hypertext. Hypermedia, 2(2): 52. Liben, L. (1981). Spatial representation and behavior: 133–158. Multiple perspectives. In: Liben, L.S., Patterson, A.M., 34. Jones, W.P., & Dumais, S. (1986). The spatial metaphor & Newcombe, N. (eds.) Spatial representation and be- for user interfaces: Experimental test of reference by havior across the life span. New York: Academic Press, location versus naming. ACM Transactions on Office pp. 16–37. Information Systems, 4:42–63. 53. Tolman, E.C. (1948). Cognitive maps in rats and men. 35. Stanton, N.A., & Baber, C. (1994). The myth of navi- Psychological Review, 55:189–208. gating in hypertext: How a “bandwagon” lost its 54. Presson, C.C., & Somerville, S.C. (1985). Beyond ego- course! Journal of Educational Multimedia and Hyper- centrism: A new look at the beginnings of spatial rep- media, 3:235–249. resentation. In: Wellman, H.M. (ed.) Children’s search- 36. Mayes, T., Kibby, M., & Anderson, T. (1990). Sign- ing: The development of search skill and spatial posts for conceptual orientation: Some requirements representation. Hillsdale, NJ: Erlbaum, pp. 1–26. for learning from hypertext. In: McLeese, R., & Green, 55. Lieblich, I., & Arbib, M. (1982). Multiple representa- C., (eds.) Hypertext: State of the art. Oxford: Intellect. tion of space underlying behaviour and associated 37. Lakoff, G. (1994). What is metaphor? In: Barnden, J.A., commentaries. The Behaviour and Brain Sciences, Holyoak, K.J., (eds.) Advances in connectionist and 5(4):627–660. neural computation theory: Analogy, metaphor and re- 56. Kaplan, S. (1973). Cognitive maps in perception and minding. New Jersey: Ablex. thought. In: Downs, R.M., Stea, D., (eds.) Image and 38. Carroll, J.M., Mack, R.L., & Kellogg, W.A. (1988). In- environment. Chicago, IL: Aldine, pp. 63–78. terface metaphors and interface design. In: Helander, 57. Golledge, R.G., & Stimson, R.J. (1997). Spatial behav- M., (ed.) Handbook of human-computer interaction. ior: A geographic perspective. New York: Guilford Press. North-Holland: Elsevier Science Publishers, pp. 67– 58. Kitchin, R.M. (1994). Cognitive maps: What are they 85. and why study them? Journal of Environmental Psy- 39. Blumenthal, L.M. (1953). Theory and applications of dis- chology, 14:1–19. tance geometry. London: Oxford. 59. Stevens, A., & Coupe, P. (1978). Distortion in judging 40. Shepard, R.N. (1962). The analysis of proximities: spatial relations. Cognitive Psychology, 10:422–437. Multidimensional scaling with an unknown distance 60. McNamara, T.P., Hardy, J.K., & Hirtle, S.C. (1989). function. Part I. Psychometrika, 27:219–246. Subjective hierarchies in spatial memory. Journal of 41. Kruskal, J.B. (1964). Multidimensional scaling by op- Experimental Psychology: Learning, Memory and Cogni- timizing goodness of fit to a nonmetric hypothesis. tion, 15:211–227. Psychometrika, 29:115–129. 61. Acredolo, L.P. (1985). Coordinating perspectives on 42. Guttman, L. (1971). Measurement as structural the- infant spatial orientation. In: Cohen, R., (ed.) The de- ory. Psychometrika, 36:465–506. velopment of spatial cognition. New Jersey: Erlbaum. 43. Carroll, J.D., & Wish, M. (1974). Multidimensional 62. Cohen, R. (1985). The development of spatial cognition. perceptual models and measurement methods. In: New Jersey: Erlbaum, pp. 115–133. Carterette, E.C., Friedman, M.P., (eds.) Handbook of 63. Siegel, A.W., & White, S.H. (1975). The development perception, Vol. 2. New York: Academic Press, pp. of spatial representation of large-scale environments. 391–447. In: Reese, H.W. (ed.) Advances in child development and 44. Tversky, A. (1977). Features of similarity. Psychologi- behavior, Vol. 10. New York: Academic, pp. 9–55. cal Review, 84:327–352. 64. Dillon, A., McKnight, C., & Richardson, J. (1990) Nav- 46 BOECHLER

igation in hypertext: A critical review of the concept. text: A meta-analysis of experimental studies. Human- In: Diaper, D., et al., (eds.) Human-Computer Interac- Computer Interaction, 11:125–156. tion—INTERACT’90. North-Holland: Elsevier Science 74. Dee-Lucas, D., & Larkin, J.H. (1995). Learning from Publishers, pp. 587–592. electronic texts: Effects of interactive overviews for in- 65. Montello, D.R. (1998). A new framework for under- formation access. Cognition and Instruction, 13(3):431– standing the acquisition of spatial knowledge in 468. large-scale environments. In: Golledge, R.G., Egen- 75. Simpson, A., & McKnight, C. (1990). Navigation in hofer, M.J. (eds.) Spatial and temporal reasoning in geo- hypertext: Structural cues and mental maps. In: graphic information systems. New York: Oxford Uni- McLeese, R., & Green, C. (eds.) Hypertext: State of the versity Press, pp. 143–154. Art. Oxford: Intellect. 66. McNamara, T.P., Halpin, J.A., & Hardy, J.K. (1992). 76. Chou, C., & Lin, H. (1997). Navigation maps in a com- Spatial and temporal contributions to the structure of puter-networked hypertext learning system. Paper pre- spatial memory. Journal of Experimental Psychology: sented at the annual meeting of the Association for Learning, Memory and Cognition, 18:555–564. Educational Communications & Technology, Albu- 67. Kuipers, B. (1982). The “map in the head” metaphor. querque, New Mexico. Environment and Behavior, 14:202–220. 77. Marchionini, G. (1989). Information-seeking strate- 68. Kosslyn, S.M. (1987). Seeing and imagining in the gies of novices using a full-text electronic encyclope- cerebral hemispheres: A computational approach. dia. Journal of the American Society for Information Sci- Psychological Review, 94:148–175. ence, 40:54–66. 69. Golledge, R.G., Ruggles, A.J., Pelligrino, J.W., & Gale, 78. Gopher, D. (1994). Dual-task performance: Rationale N.D. (1993). Integrating route knowledge in an unfa- and scope. In: Eysenck, M.W., (ed.) The Blackwell dic- miliar neighborhood: Along and across route experi- tionary of cognitive psychology. Cambridge, MA: Black- ments. Journal of Environmental Psychology, 13:293–307. well, pp. 111–116. 70. Hirtle, S.C., & Hudson, J. (1991). Acquisition of spa- 79. Ericcson, K.A., & Simon, H.A. (1984). Protocol analy- tial knowledge for routes. Journal of Environmental sis: Verbal reports as data. Cambridge, MA: MIT Press. Psychology, 11:335–345. 71. Garling, T., Book, A., & Lindberg, E. (1989). Adults’ memory representations of the spatial properties of Address reprint requests to: their everyday physical environment. In: Cohen, R., (ed.) The development of spatial cognition. New Jersey: Patricia M. Boechler Erlbaum, pp. 141–177. Department of Psychology 72. Korthauer, R.D., & Koubek, R.J. (1994). An empirical P108 Biological Sciences Bldg. evaluation of knowledge, cognitive style, and struc- University of Alberta ture upon the performance of a hypertext task. Inter- Edmonton, Alberta, Canada T6G 2E9 national Journal of Human-Computer Interaction, 6(4): 373–390. 73. Chen, C., & Rada, R. (1996). Interacting with hyper- E-mail: [email protected]